While it’s true that the AI language models that power chatbots are trained on terabytes of text, the memory in these systems is limited. Founded by former OpenAI engineers, Anthropic has greatly expanded the context window for its chatbot named Claude to 75,000 words. In ChatGPT, that’s about 3,000 words.Thus, Claude, developed as a competitor to Bard, Bing and ChatGPT, was able to fully understand a classic novel in just 30 seconds. Claude’s new capability is currently only available to Anthropic partners who connect to the chatbot via the company’s API.
While processing large volumes of text requires significant computational resources, the increased capability of AI language models will certainly make these systems more useful. Chatbots’ context window is measured against tokens that don’t exactly correspond to words or characters.
He found the change made in the novel in 22 seconds
Claude could previously process around 9,000 tokens. Now it can process 100,000 tokens, or about 75,000 words. For comparison, OpenAI’s GPT-4 can handle around 8,000 tokens (we’re talking about a paid service), while the full version of GPT-4 can handle up to 32,000.
Anthropic highlights that it takes about five hours for a human to read a 75,000-word text, but with Claude’s expanded context window, he can summarize and analyze long documents in a matter of minutes. The developers asked Claude to read the Great Gatsby novel to find the change they had made to a line. The chatbot correctly detected the change in 22 seconds.
57606
